A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering, School of Advanced Technologies

Thesis (M.Eng.) - Asian Institute of Technology, 2005

In medicine, anatomical sketches are omnipresent. Physicians frequently draw sketches in patient records and use them to communicate with either patients or colleagues regarding diagnoses and treatments. In medical education, students often use sketches for learning anatomy as well as communicating in collaborative problem solving. If a computer system is capable of understanding anatomical structures sketched by humans, various medical applications such as electronic patient records and medical tutoring systems could be enhanced especially in an aspect of intelligent interfaces. In this thesis, we proposed an approach to segment detailed structures of anatomical sketches (e.g. parts of the heart) even if some parts have not been explicitly drawn. The proposed segmentation algorithm consists of five major processes; point matching, distance-based elimination, transformation, connecting boundary points, and labeling interior points. An empirical analysis presented in this study explores the agreement between the segmentations of the system and those of an experienced physician. In addition, two subjective evaluations are also conducted: senior medical students' judgments of the quality of the system's segmentations and an experienced physician's segmentations, and a survey to determine a suitable parameter to apply in the algorithm. Finally, a prototype system is implemented to demonstrate the capability of the anatomical sketch segmentation in practice